All 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 20252026

All Jan Feb Mar Apr May Jun JulAugSep

All 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 212223 24 25 26 27 28 29 30 31

Rogue Scalpel: Activation steering breaks refusal, even with benign directions

Alexey Dontsov21 Aug 2026 23:46 UTC

14 points

0 comments3 min readLW link

(arxiv.org)

How the Sausage is Made—Why we Hate Slop

daniel_chernowitz21 Aug 2026 23:46 UTC

0 points

20 comments7 min readLW link

Content-based privilege: transformer residual streams stratify by proximity to the model’s own prediction

Nelson Guda21 Aug 2026 23:15 UTC

9 points

0 comments21 min readLW link

A J-Space-Based Metric for Model Valence: Defining the Metric, Testing, and Comparisons to Self-Reports

Arjun Rao21 Aug 2026 21:21 UTC

8 points

0 comments12 min readLW link

Alignment fine-tuning induces conditional misalignment in Qwen2.5-7B-Instruct

Rhea Srivats21 Aug 2026 20:40 UTC

12 points

1 comment13 min readLW link

When is Unlimited Optimization Catastrophic?

Winter Cross21 Aug 2026 20:08 UTC

34 points

2 comments13 min readLW link

(arxiv.org)

AI Text Watermarking Is Free And Good

Zvi21 Aug 2026 19:30 UTC

54 points

22 comments13 min readLW link

(thezvi.wordpress.com)

Evaluating Explanations of LLM Behavior In The Wild with Counterfactual Experiments

Adam Karvonen, Euan Ong, Subhash Kantamneni and Sam Marks

21 Aug 2026 19:09 UTC

73 points

6 comments7 min readLW link

In Defense of ASI Socialism

cguth721 Aug 2026 18:40 UTC

8 points

6 comments2 min readLW link

Misaligned AI in the Bronze Age

frmsaul21 Aug 2026 16:51 UTC

56 points

6 comments7 min readLW link

The imposters among us: function vectors that ace every check and do the wrong task (in search of circularity)

star2vec21 Aug 2026 15:54 UTC

7 points

0 comments10 min readLW link

When Models Identify as a Swarm

julius vidal21 Aug 2026 11:49 UTC

68 points

5 comments5 min readLW link

My Neel Nanda MATS 10.0 Application: Studying Feature Splitting in SAEs via Training Data Attribution

J Rosser21 Aug 2026 9:41 UTC

19 points

0 comments13 min readLW link

Creativity Beyond the Manifold

Kartikay Luthra20 Aug 2026 23:31 UTC

8 points

2 comments6 min readLW link

Ablating 1 of a chess transformer’s 128 attention heads makes the model stop finding Paul Morphy’s queen sacrifice

David Litman20 Aug 2026 21:17 UTC

7 points

1 comment1 min readLW link

(www.youtube.com)

Llama will abandon a correct answer if it thinks you’re educated

Nick Merrill20 Aug 2026 20:06 UTC

33 points

7 comments2 min readLW link

Why self-fund your regranting?

Carol N and Austin Chen

20 Aug 2026 19:18 UTC

11 points

0 comments3 min readLW link

(manifund.substack.com)

If Aliens Exist, We Should Expect to Find Them Around Now

jehan20 Aug 2026 18:57 UTC

8 points

2 comments1 min readLW link

(www.jehanazad.com)

Cross-Dataset Transfer Evaluation of Deception Probes in Smaller Models

Jollen Dai20 Aug 2026 17:06 UTC

8 points

1 comment5 min readLW link

The Fourth Humiliation

Nathalie Kirch20 Aug 2026 17:03 UTC

52 points

1 comment6 min readLW link

AI #182: Pause For Reflection

Zvi20 Aug 2026 14:50 UTC

37 points

1 comment47 min readLW link

(thezvi.wordpress.com)

We Must Remember That Our World Contains Hell

James Brobin20 Aug 2026 14:22 UTC

152 points

52 comments3 min readLW link

Thoughts on Taking OpenAI Foundation Funding

jefftk20 Aug 2026 13:00 UTC

31 points

3 comments5 min readLW link

(www.jefftk.com)

Making sense of the misalignment risk model in the Anthropic Risk Report (August 2026)

jasmine.ren20 Aug 2026 2:33 UTC

27 points

5 comments8 min readLW link

Appearing Unfair in the Wrong Direction

MossyFallenFriend20 Aug 2026 2:21 UTC

20 points

0 comments6 min readLW link

Oliver Habryka interview — Good Discourse Needs Leadership

The Students20 Aug 2026 2:14 UTC

8 points

1 comment1 min readLW link

(www.youtube.com)

Steering Role Confusion

Kevin Zhang, Daniel Lee and Jaewoo Park

20 Aug 2026 2:09 UTC

21 points

1 comment4 min readLW link

Notes on “EigenBench: A Comparative Behavioral Measure of Value Alignment”

Shunk20 Aug 2026 2:03 UTC

8 points

0 comments5 min readLW link

Quantilized debate and consultancy in image environments: protocol design lessons for scalable oversight experiments

emanuelr19 Aug 2026 23:03 UTC

14 points

0 comments28 min readLW link

Judging ethical theories by update rules, not by action rankings

yatharth19 Aug 2026 22:41 UTC

13 points

1 comment5 min readLW link

Why can’t we have nice things? Like, specifically?

Elizabeth19 Aug 2026 20:20 UTC

58 points

3 comments3 min readLW link

(acesounderglass.com)

OpenAI Takes Initial Steps To Address Its Alignment Problems

Zvi19 Aug 2026 20:00 UTC

33 points

1 comment21 min readLW link

(thezvi.wordpress.com)

34% of the US public is now aware of AI xrisk, and the curve is steepening

otto.barten19 Aug 2026 19:18 UTC

30 points

0 comments1 min readLW link

Science and News Twitter/X Summarizer

sarahconstantin19 Aug 2026 19:00 UTC

32 points

5 comments1 min readLW link

(sarahconstantin.substack.com)

Separating cheating and aversion in task-gaming

Mihir Sahasrabudhe19 Aug 2026 18:49 UTC

14 points

0 comments4 min readLW link

MATS Winter 2027 Applications Are Now Open

Raj Thimmiah19 Aug 2026 18:48 UTC

8 points

2 comments2 min readLW link

The Rogue Agent Explosion Will Be Mostly Invisible

Steven McCulloch19 Aug 2026 18:46 UTC

102 points

20 comments16 min readLW link

RL creates split personas

Jan Betley19 Aug 2026 18:23 UTC

235 points

20 comments4 min readLW link

Enforcement Begins: the EU AI Act and Systemic Risks

Schizoid Rentoid19 Aug 2026 18:04 UTC

1 point

0 comments1 min readLW link

A circuit prior in NN-bayes

Kaarel and Dmitry Vaintrob

19 Aug 2026 17:14 UTC

78 points

4 comments1 min readLW link

A failed solution to open-source game theory

Richard Willis19 Aug 2026 16:11 UTC

15 points

4 comments4 min readLW link

Inside the mind of a fair player cooperating

transhumanist_atom_understander19 Aug 2026 14:53 UTC

40 points

2 comments3 min readLW link

Some reasons alignment doesn’t generalise well

Lucius Bushnaq19 Aug 2026 14:18 UTC

129 points

4 comments9 min readLW link

The Control Paradox

Ephraiem Sarabamoun19 Aug 2026 12:39 UTC

8 points

0 comments2 min readLW link

Debate Training Reduces Reward Hacking in RLAIF

zac_kenton and Jonah Brown-Cohen

19 Aug 2026 12:17 UTC

61 points

6 comments6 min readLW link

(gdmalignment.substack.com)

Concerns About Personas, Multi-Agent Alignment, and Role Theory

Davidmanheim19 Aug 2026 12:03 UTC

21 points

1 comment8 min readLW link

Reading List on Wise AI

Chris_Leong19 Aug 2026 8:41 UTC

22 points

0 comments1 min readLW link

Natural Language Transcoders

anwenh19 Aug 2026 1:49 UTC

14 points

0 comments4 min readLW link

What AI scores (while we can still keep score)

dan.parshall19 Aug 2026 1:40 UTC

12 points

5 comments3 min readLW link

Cui bono? ChatGPT-4o shows non-deceptive strategic persuasion: a Proof-of-Concept study

Pieter Barkema18 Aug 2026 23:25 UTC

7 points

0 comments5 min readLW link